@davoxi/mcp-server

MCP (Model Context Protocol) server for the Davoxi AI voice agent platform. Lets AI assistants manage voice businesses, specialist agents, call logs, webhooks, usage analytics, billing, and API keys conversationally.

Works with Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible client.

Installation

Option 1: npx (recommended, zero install)

No installation needed. Just add the config to your AI tool and it runs automatically.

Option 2: Global install

npm install -g @davoxi/mcp-server

Option 3: From source

git clone https://github.com/MabrokaMedia/davoxi-mcp-server.git
cd davoxi-mcp-server
npm install && npm run build

Related MCP server: ElevenLabs MCP Server

Authentication

Browser Login (recommended)

The easiest way to authenticate — no API key copy-paste needed:

npx @davoxi/mcp-server auth login

This opens your browser to the Davoxi dashboard where you log in and click "Authorize". The API key is created automatically and saved to ~/.davoxi/mcp.json.

Other auth commands:

npx @davoxi/mcp-server auth status   # Check current auth state
npx @davoxi/mcp-server auth logout   # Clear saved credentials

Manual API Key

Alternatively, get an API key from your Davoxi Dashboard and set it as an environment variable.

Key resolution order: DAVOXI_API_KEY env var > ~/.davoxi/mcp.json (from browser login)

Connecting to Your AI Tool

Prerequisites

1. Authenticate: Run npx @davoxi/mcp-server auth login (or get an API key manually)

2. Node.js 20+ installed on your machine

Claude Code

Run this command in your terminal:

claude mcp add davoxi -- npx -y @davoxi/mcp-server

Then set your API key in the environment. Or add to .mcp.json in your project:

{
  "mcpServers": {
    "davoxi": {
      "command": "npx",
      "args": ["-y", "@davoxi/mcp-server"],
      "env": {
        "DAVOXI_API_KEY": "sk_your_key_here"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "davoxi": {
      "command": "npx",
      "args": ["-y", "@davoxi/mcp-server"],
      "env": {
        "DAVOXI_API_KEY": "sk_your_key_here"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "davoxi": {
      "command": "npx",
      "args": ["-y", "@davoxi/mcp-server"],
      "env": {
        "DAVOXI_API_KEY": "sk_your_key_here"
      }
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "davoxi": {
      "command": "npx",
      "args": ["-y", "@davoxi/mcp-server"],
      "env": {
        "DAVOXI_API_KEY": "sk_your_key_here"
      }
    }
  }
}

Environment Variables

Available Tools (27)

Businesses (5 tools)

Agents (7 tools)

Call Logs (2 tools)

Agent Testing (1 tool)

Webhooks (4 tools)

Phone Numbers (1 tool)

Analytics & Billing (4 tools)

Account (4 tools)

Example Workflows

Set up a new business with agents

You: Create a business called "Acme Support" with voice "nova", language "en-US",
     and personality "You are a friendly, professional customer service agent."

You: Create an agent for that business:
     - Description: "Handles appointment scheduling"
     - System prompt: "You help callers schedule, reschedule, or cancel appointments.
       Always confirm the date, time, and service before booking."
     - Trigger tags: ["appointment", "schedule", "booking", "reschedule", "cancel"]

You: Create another agent for billing questions with trigger tags ["billing", "invoice", "payment"]

You: List all agents for Acme Support to verify everything looks good

Monitor call activity

You: Show me all calls for business biz_abc123 from the last 7 days

You: How many calls were missed vs completed this week?

You: Show me the details and transcript for call call_xyz789

Set up webhook notifications

You: Create a webhook for business biz_abc123 that sends to
     https://hooks.example.com/davoxi with events: call.completed, call.missed

You: List all webhooks for that business to verify

Test an agent end-to-end without WhatsApp

You: I just deployed agent agent_xyz on biz_abc123 — try asking it
     "what's your weekend availability?" and tell me what it replies

You: Now follow up with "great, can I book Saturday at 2pm?" using the
     same caller_id so it remembers the prior turn

test_agent_chat runs through the same Brain → Specialist chain that real Twilio traffic does, so the reply you get back is exactly what a real caller would receive. The Brain may emit a filler ("one moment please…") followed by the Specialist's final answer — both come back in the replies array, in order.

Duplicate and modify an agent

You: Duplicate the appointment booking agent from Acme Support

You: Update the copy's system prompt to handle Spanish-speaking callers
     and change trigger tags to ["cita", "reservar", "cancelar"]

You: Enable the new agent

Security

Authentication

• All API calls require a valid DAVOXI_API_KEY (Bearer token)

• API keys are scoped to your account — they can only access your businesses, agents, and data

• Keys start with sk_ prefix for easy identification

• Full API key is only shown once at creation — store it securely

Browser Login Security

• CSRF protection — random 64-character state parameter validated on callback

• Localhost callback — the temporary server binds to 127.0.0.1 only (unreachable from network)

• Single-use — callback server closes immediately after receiving one response

• 5-minute timeout — abandoned login flows are automatically cleaned up

• File permissions — ~/.davoxi/mcp.json created with mode 0o600 (owner read/write only)

• No open redirect — login redirect validated to start with /

Transport Security

• HTTPS enforced — the server validates that DAVOXI_API_URL uses HTTPS (http://localhost allowed for development only)

• Stdio transport — the MCP server communicates via stdin/stdout, not over the network. Your API key never leaves your machine except in authenticated API calls to Davoxi

• No open ports — the server doesn't listen on any network port

Data Protection

• API key prefixes are returned for identification, but full keys are never re-exposed after creation

• SSRF protection — agent tool endpoints are validated to prevent calls to private/internal IPs (127., 10., 192.168., 169.254., etc.)

• IPv6 private ranges blocked (::1, fe80:, fc00:, fd00:)

• Blocked hostnames: localhost, ip6-localhost (in tool endpoint validation)

Destructive Operation Safety

All destructive operations require explicit confirm=true parameter:

• delete_business — warns about cascading deletion of agents, phone assignments, and webhooks

• delete_agent — suggests disabling instead (enabled=false is reversible)

• revoke_api_key — warns about immediate access loss for integrations

• delete_webhook — confirms permanent removal

Best Practices

1. Use separate API keys for different environments (production, staging, development)

2. Name your API keys (create_api_key with name parameter) for easy identification

3. Rotate keys regularly — create a new key, update integrations, then revoke the old one

4. Use webhooks over polling — subscribe to events instead of repeatedly calling list_call_logs

5. Disable before deleting — use update_agent with enabled=false before permanently deleting

Webhook Security

• Webhook responses include a secret field at creation — use it to verify payload signatures

• Endpoints must respond with 2xx within 10 seconds

• Failed deliveries are retried up to 3 times with exponential backoff

• Only HTTPS webhook URLs are accepted

Development

npm run dev    # Watch mode (recompiles on change)
npm run build  # Production build
npm start      # Run the server
npm test       # Run tests

License

MIT